Northbridge + Southbridge = Unitedbridge? Is the division likely to persist?

Question

Okay, hardware based question.

I was at shopping for laptops the other day and one of the salespeople was trying to talk up an integrated graphics system as opposed to a discrete graphics system. He said Intel and other chip fab/designers were trying to do away with the Northbridge VS Southbridge design concepts. Now I have a degree in IS and Finance so this comment seemed a little counterintuitive to me.

The isolation of Southbridge components from Northbridge components is a logical design convention so as to separate to quick components (RAM, CPU, AGP) from the slow components (HDD, Optical, Tape, USB, etc). Why would the system designers try to integrate these divided components? It seems like a step backwards to me; even with a system on chip (SoC) like the Tegra, the components that are included in the SoC are just tightly integrated Northbridge functions, right?

I suppose its possible that designers/fabricators want “unitedbridge” type system to allow for a more uniform fabrication and programming environment. Limit the number of hardware configurations out there and program development might become an easier endeavor. Isn’t that why people are so hot on ARM/RISC development right now?

Was the salesperson just full of hot air or do I need to adjust my news feeds?

Answer 1

Without delving into too much details, it is unlikely that the north and southbridges will be combined together in the near future. They are two significantly different buses, and were separated for one reason - speed.

Most devices on the southbridge aren't bandwidth intensive, and don't require high speeds (e.g. USB devices, network cards, sound cards). In contrast, the northbridge implements a high-bandwidth, low-latency interface between the CPU, RAM, and GPU in the system (since these are very bandwidth intensive components).

While the two may not be merged anytime in the near future, newer CPUs are actually implementing some northbridge functions directly on the processor die. This is especially the case with the memory controller, which provides direct CPU control over the memory, reducing latency.

Answer 2

I don't see this happening except for form factor computers.

ARM is popular because the mobile world is growing. However, ARM sucks on desktop/server hardware. Try to upsize an ARM to compete with an i7 and it will get whipped like a schoolboy. On top of that, ARM loses it's considerable power advantage at that size and workload. It's the classic scaling fallacy (aka cube law, aka law of sizes) that gets people.

This forces the two designs into the same roadmap. Traditionally, the northbridge needed more iterations to keep up with the breakneck pace of processor upgrade.

Second, it screws with chip manufacturer's ability to market to a wider audience. Modulations of your product actually makes it easier to market to multiple audiences. Most businesses actually want to have multiple hardware configurations.

If you did push them into one package, it would also increase crosstalk and heat, furthering problems. And more than likely, it will just be two chips in one package to further binning.

This salesperson is being way to narrow minded IMO.